Zhang, Bin; PhD

                         DUKE UNIVERSITY, 1999

                         ECONOMICS, GENERAL (0501); ECONOMICS, FINANCE (0508)
 

                         This dissertation investigates the properties of dynamic model selection procedures in empirical analysis
                         without reliance on a specific class of models. In the first chapter, probability convergence rates of the
                         dimension estimators are derived using asymptotic approximates. We present the convergence results
                         of model selection procedures for cases with and without model-class misspecifications. Furthermore, a
                         robust consistent procedure is proposed to determine the specifications of more complicated models,
                         such as continuous-time diffusion models, when the likelihood function is intractable. In the second
                         chapter, Monte Carlo experiments are conducted to demonstrate the finite sample performances of
                         various information criteria including BIC and PIC, and to corroborate the analytical probability
                         convergence rates. To achieve these goals, we investigated a very comprehensive list of models by
                         carefully controlling the key factors that can make our conclusions more scientific. The Monte Carlo
                         results show that the approximated probability convergence rates are close representations of the
                         empirical frequencies of misspecification in the dynamic model selection process.

Social Systems Simulation Group P.O. Box 6904 San Diego, CA  92166-0904 Roland Werner, Principal Phone/FAX  (619) 660-1603